PROJECT SUMMARY/ABSTRACT My goal is to pursue an independent career in computational psychiatry by leveraging cutting-edge neuroimaging and data-driven analysis approaches to advance precision medicine in mental health. To build on my strong neuroimaging and computational background, the training component of this award emphasizes coursework and mentorship in the clinical and behavioral aspects of psychopathology. I will also receive mentorship to advance my theoretical and applied understanding of deep learning in this burgeoning field. The overarching research goal in this proposal is to develop computational strategies that account for the heterogeneity of mood disorders to improve the identification of treatment-response biomarkers. Response to pharmaceutical and behavioral antidepressant treatments is low, likely due to the symptomatic and etiological heterogeneity of depression whereby certain treatments may confer differential benefits for patients having particular symptom constellations. In the K99 phase, I will seek to improve prediction of individual antidepressant response using electroconvulsive therapy (ECT), which elicits robust and rapid antidepressant effects, as the treatment model. I will use MRI and clinical data from patients undergoing ECT collected for the large the Global ECT-MRI Research Collaboration (GEMRIC). In Aim 1, I will use exploratory factor analysis to characterize latent symptom dimensions of the GEMRIC cohort before, during, and after ECT. The accuracy of predicting clinical outcomes along the recovered symptom dimensions will be compared to traditional means of evaluating response using the total score of the Hamilton Depression Rating Scale (HDRS). Pursuit of this aim will expand my understanding of clinical psychiatry and lay foundational knowledge for the independent aims. Aim 2 will expand my deep learning and multimodal neuroimaging skillsets as I develop novel deep learning architectures to fuse multimodal imaging features of GEMRIC participants to further improve predictions of treatment response and cognitive impairment following ECT. Rather than simply concatenating multimodal features together, deep network architectures will discover latent feature representations. The R00 phase will be a logical progression of the skill sets I develop in the mentored phase and expand on these lines of research. Aim 3 will draw from a collection of large-scale MRI datasets from patients with more broadly defined mood disorders to identify multimodal imaging markers associated with transdiagnostic symptom domains. Aim 4 uses treatment groups from aim 3, including patients undergoing ketamine, sleep deprivation, cognitive behavioral therapy, and pharmaceuticals, to explore the extent to which biomarkers of therapeutic response, defined along the transdiagnostic symptom dimensions identified in Aim 3, are shared across treatment groups. I anticipate that discrete categorizations of mood disorders artificially obscures discovery of treatment-response biomarkers. Fulfillment of these aims will simultaneously propel me to independence and yield important insight into the treatment of heterogeneous mood disorders.